Content
75%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A dense, well-organized single-file reference whose value is in genuinely specific recommended values (JWT windows, header strings, severity actions) and a real evidence-producing verification checklist. Its main gaps are the absent explicit review sequence, the unprovided grep patterns it asks the agent to show, and one dangling file reference.
Suggestions
Provide the concrete grep/ripgrep patterns behind 'show grep results for common secret patterns' so the verification step is self-contained rather than requiring the agent to invent patterns.
Add a short 3-4 step 'How to run a security review' sequence (scan Red Flags, apply OWASP table, assign P0-P3 severity, run Verification checklist) to make the currently implicit workflow explicit.
Remove or resolve the reference to 'archived-agent-rejection.md §C', which is not part of the skill bundle and reads as a dangling pointer to readers without that parent context.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense reference tables with almost no prose padding, and most rows carry specific values Claude cannot be trusted to recall from memory ('bcrypt (cost factor 12+) or Argon2id', 'Access Token Expiry | 15-30 minutes', 'max-age=31536000; includeSubDomains'). Held below 5 because several rows restate well-known basics ('SQL Injection | Parameterized queries', 'XSS Prevention | HTML escaping') and the 'Common Rationalizations' section is persuasive padding that assumes the agent needs convincing rather than reference data. | 4 / 5 |
Actionability | Concrete throughout: exact settings ('RS256 (asymmetric)', 'httpOnly + secure + sameSite cookie'), named tools per task ('Zod, Joi, pydantic, Go validator', 'DOMPurify (client)', 'Prisma, GORM, SQLAlchemy'), and severity-gated actions ('P0 | CRITICAL | Block release'). Not 5 because the verification checklist demands 'show grep results for common secret patterns' without providing the patterns themselves, and tool mentions lack even minimal usage snippets — minor but real gaps in an otherwise executable reference. | 4 / 5 |
Workflow Clarity | The 'Verification' section supplies explicit checkpoints with evidence requirements ('show which items were evaluated', 'show grep results for common secret patterns') and 'Security Review Severity Levels' maps each finding to a concrete action. However, no explicit ordering of a review exists — how the Red Flags, OWASP table, severity levels, and Verification sections chain together is left implicit, so it fits the 'clear structure, most checkpoints present, minor gaps' anchor rather than the full sequence-plus-feedback-loop anchor. | 4 / 5 |
Progressive Disclosure | A single-file skill with no references/, scripts/, or assets/ directories; sections are clearly headed and the ~130-line body of compact tables is appropriately kept in one file rather than split. Not 5: it exceeds the compact under-50-line case where sections alone suffice, and the body cites 'archived-agent-rejection.md §C' — a document not present in the bundle — a dangling pointer that slightly undermines navigation. | 4 / 5 |
Total | 16 / 20 Passed |